2014 - Round 21 Results : A Welcome Return to Form

After two successive weeks of losses, both the Head-to-Head and Line Funds returned to profitability this week, the Head-to-Head Fund landing its sole bet on Freo to record a 6c profit and the Line Fund guessing correctly on four of five occasions to record a profit of about the same size.

Those two profits, paired with the Margin Fund's now-customary loss - this week of 6.5c despite a successful call in the Dons v Eagles game - lifted the Recommended Portfolio by 3.7c for the week, leaving it now up by 7.7c on the season.

Absent a stunning bout of prescience on the part of the Margin Fund, I think it's safe to suggest that this Fund will now almost certainly be retiring at the end of this season, at least in its current form.

It could though have been a very different story for this Fund had it not based its wagering activity this year on Combo_NN_2 and Bookie_9, who have proven to be among the worst bucket predictors this season.

Basing its wagering instead on any of the Head-to-Head Predictors or on either of the ProPred Predictors, while retaining its habit of wagering only on home team wins or draws, would have yielded the Margin Fund ROIs in the 15 to 30% range. Even using either of the WinPred Predictors would have yielded a small profit. These Predictors have all selected the correct bucket about twice as often as Combo_NN_2, which has been right only seven times all season, and about 50% more often than Bookie_9, which has been bucket-perfect on just nine occasions.

(It's interesting to note that the two WinPred Predictors returned a profit last year too when wagering only on their home team win or draw predictions. It's tempting to think about switching to them for the Margin Fund for next season, but that does feel a bit like making your Lotto picks based on last week's results ...)


There were, you might recall, high levels of disagreement amongst the Head-to-Head Tipsters this week, and this served to widen the span of scoring. Best was Silhouette's 8 from 9, and worst was Easily Impressed's 3 from 9.

Nearer the top of the the Leaderboard some more important differences in performance were seen for the three Win-based Predictors, who all scored 7 from 9 to move them into equal third, and for Combo_NN_1, whose 4 from 9 dropped it into a tie for eighth with C_Marg.

The All-Tipster average for the round was 6 from 9.

The Margin Predictors mean Mean Absolute Prediction Error (MAPE) for the week was 31.5 points per game, made smaller by the MAPEs of the four H2H-based Predictors of around 28.7 to 28.9 points per game, and enlarged by the Win-based Predictors MAPEs of 34.8 and 35.3 points per game, and by Combo_NN_1's 36.6 points per game.

Combo_7's MAPE of 29.9 was sufficient for it to retain the outright lead while Bookie_LPSO's 30.8 and Bookie_9's 30.6 were good enough, relative to Win_3's to elevate them into second and third positions.

Last year's best Margin Predictors, RSMP_Simple and RSMP_Weighted, are now close enough to make back-to-back top 2 performances a possibility.

Thirteen of the Margin Predictors now have Line Betting records better than 52.6%, which makes them sufficiently accurate (at a $1.90 price per wager) to be profitable.

The round's best Head-to-Head Probability score belonged to the two H2H algorithms. They, along with Bookie-OE, were the only Probability Predictors to return positive scores for the round. C_Prob recorded the round's worst result and, consequently, dropped to fourth on the Leaderboard.

MoS' Line Fund algorithm earned its best probability score since Round 15, snapping a five-week stretch of negative scores.